LECTURE 5 MORAL AGENCY AND MORAL RESPONSIBILITY LARS

LECTURE 5: MORAL AGENCY AND MORAL RESPONSIBILITY LARS LINDBLOM CTE LARS. LINDBLOM@LIU. SE

WHY ETHICS IN A COURSE ON PROFESSIONALISM IN COMPUTER SCIENCE? • Some general answers • Better at what you do: identifying value conflicts • Better understanding of what you do: analyzing value conflicts • Better explaining what you do: working with others • A more specific answer: Professionalism is (in part) an ethical ideal • Todays topics • What is ethics? • Ethical theory as an aid to engineers • Responsibility and engineering

RATIONALITY IN RUSSEL AND NORVIG’S ARTIFICIAL INTELLIGENCE – A MODERN APPROACH • “A rational agent is one that does the right thing. ”(p. 36) • The definition of a rational agent: “For each possible percept sequence, a rational agent should select an action that is expected to maximize its performance measure, given the evidence provided by the percept sequence and whatever built in knowledge the agent has. ” (p. 37). • This is standard rational choice theory: Maximizing expected utility • The same theory is used in economics, political science, psychology evolutionary biology, law and philosophy • Desirability in standard theory is usually spelled out in terms of preferences but here the term performance measure takes its place. • There has been a lot of discussion in other fields of this theory, and that discussion seems relevant for theory of AI

RATIONALITY AND IMPARTIALITY • A common intuition: morality has to do with impartiality • Rational choice theory is not necessarily impartial • A controversial aspect of economic theory • Economist and Nobel Laureate John Harsanyi: The theory of rationality should work in both positive and normative roles • Solution: Ethics has to do with being rational behind a veil of ignorance • Equal probability to be each person • The impartial observer will choose as if s/he were an average utilitarian • Rationality and impartiality is compatible

ETHICS: FUNDAMENTAL CONCEPTS • Ethics is the study of morality (but basically the same word) • Descriptive Ethics: How people act and think about morality • Normative Ethics: How people should act and think. . • Normative judgement: Judgement about whether something is good or bad, desirable or undersirable, right or wrong. • Hume’s Law: You cannot derive ought from is • Values: To be strived for • • Intrinsic and Instrumental • The example of privacy Norms: Rules tha prescribe what actions are required, permitted, or forbidden • • Virtues: Good permantent characteristics in a human being • • A way to realize values Justice, courage, wisdom etc. Relativism: Gotterbarn ”no matter how they shake hands, they do testing”

UTILITARIANISM • Maximize the good • The sole focus is on consequeces (consequentialism) • The good (utility) is happiness • Impartiality and hedonism implies utilitarianism • Act and Rule Utilitarianism • The Harm Principle: The only reason that justifies restrictions on freedom is to avoid harm to others • Problems • Distributive justice • Relations • Scape goats

PROMOTING AND RESPECTING VALUES • Kant vs. the utilitarians • Promoting a value: maximize the amount of that value • Respecting a value: letting it set limits to the options • Human dignity • Maximizing human dignity seems to imply that we should create more people • Respect for human dignity implies that the focus should be on the people that exist • This indicates that rational choice theory does not contain the materials to capture our intuitions about all kinds of value; maximization is not enough • Maybe AI must be rational in a wider sense than that of rational choice theory

DEONTOLOGY • Focuses on norms not consequences • The good will, the only thing that is unconditionanlly good • Two kinds of imperatives: Hypothethical and Categorical • The Cateogrical Imperative 1: ”Act only on that maxim (norm) which you can at the sam time will that it should become universal law” • • The Categorical Imperative 2: ”Act as to treat humanity, whether in your own person or in that of any other, in every case as an end, never as a means only” • • Theft and lying Ford Pinto and suicide Problems • How to handle conflicts: prima facie norms • Absolutism; the consequences never trump rights

AI, MONISM AND PLURALISM • Kant and animals • If we do not owe animals respect, then we can use them for our own purposes, and if it does not make us bad people to harm them, we may do so. • But there might be more than two kinds of value: consideration? • Monism and pluralism: is there one or many kinds of values? • Many kinds of values: welfare, beauty, nature, love, privacy, etc. • Many kinds of attitudes to value: use, respect, , consideration, , appreciation, love etc. • Impossible to choose? • Interpretation and frames: We have to know what kind of situation we are in. • Am I acting as a parent or as a businessperson? • The informational difference between science and moral choice • Simplifications are essential in science, but may be misleading for moral purposes

VIRTUE ETHICS • Foucuses on who to be, rather than consequences and norms • Virtue: Good and permanent characteristics in a human being • Something we can learn, habit is important • The virtues are the middle way between extremes • Classical virtues: prudence, temparence, justice, fortitude, and faith, hope, charity • Practical wisdom is needed to apply the virtues • Role virtues: what is an engineer supposed to be? • Problems: • What is the justification for the virtues? • How to handle conflicts between virtues? • Is theory action-guiding?

THE VIRTUES OF THE ENGINEER • Expertise/professionalism • Clear and informative communication • Cooperation • Willingness to make compromises • Objectivity • Being open to criticism • Stamina • Creativity • Striving for quality • Having an eye for detail • Being in the habit of reporting on your work carefully

THE PEAS MODEL AND ETHICS PEAS Performan Environm Actuators Sensors ce Meaent sure Agent type: Safe, Fast, Taxi Driver Legal, Maximize Profits, Comfortabl e Roads, Traffic, Walkers, Customers Steering, Camera, Accelerator, Sonar, Brake, Speedo. Signal, meter, Horn, GPS, etc. Display • To rationally strive to satisfy the performance measure given the environement • Two upshots: • PEAS is also a useful tool to analyze ethical problems • The model can also be used to illustrate the analytical tools of ethical theory

PERFORMANCE MEASURE • What is valuable, , i. e. , in taxi services • Intrinsic or instrumental value • Monism or Pluralism • Attitudes to value: Promote or respect • A common notion of intrinsic value: Welfare • Hedonism, preferentialism and objective list • Antropocentrism and biocentrism

ENVIRONMENT • How should the world be understood? • An exemple: The philosophy of social science • Are there, fundamentally, groups, and if so, do they have interests? • Metodological individualism or collektivism? • Political philosophy: How should the environment be designed? • The Bias Problem: “diskriminatory” AI depends in some degree to how the housing market connects to the banking sector.

SENSORS • Meta ethics: What is it that we do when we see that something is valuable? • Observe a fact: Realism • Express a feeling: Emotivism • An example for the philosophy of education: Should lawyers read novels a a part of their education? • No, it is a waste of time • Yes, it increases the ability to identify what is really at stake • Tom Robinson in To Kill a Mockingbird runs away, but why?

ACTUATORS • What actions are relevant? • Creativity, see Sensors • Are all actions relevant? • A core issue in normative theory • The debate between utilitarianism and kantianism • Applied Ethics: What does normative theory apply for practice; or How should we solve or problems? • Engineering ethics is a subfield of applied ethics and it focuses on responsibility

THE NOTION OF RESPONSIBILITY • Everybody seems to agree that engineers have responsibilities, from engineers to the general public, but what does this mean? • Active Responsibilty: Before something happens • Passive Responsibility: After something happens • Accountability: Having to account for one’s actions • Might show that one was not blameworthy • Role Responsiblity: One’s responsibility as, e. g. , a computer scientist • Professional responsibility • Moral Responsibility: One’s responsibility as a person

BLAMEWORTHINESS • Wrong-doing: A violation of a norm • Legal, organizational or moral • Causal Connection: (A part of) bringing about the violation • Acts and omissions • Foreseeability: Information • Ignorance may be an excuse • Freedom: Not forced • Very politically controversial what a forcing is

PROFESSIONAL IDEALS • Active responsibility is forward-looking, what are its ideals? ? • Technological Enthusiasm: Developing new technological opportunities • Morally neutaral (Wernher von Braun) • Effectiveness and Efficiency: To make better functioning artifacts • Effectivness: The extent to which an estabilshed goal is acheived • Efficieny: The ratio between the goal acheived and the effort required • The moral value of both depend on the goals promoted • Human Welfare: ”engineers shall use their knowledge and skill for the enhancement of human welfare” (American Society of Civil Engineers) • Seemingly a moral goal

THE RESPONSIBILITIES OF ENGINEERS • As a professional • As an employee • As a person • Separatism: The engineer only gives technical input • • Tri-partite model: politics, management, engineer • Hired guns? Technocracy: Government by experts • Technological expertise is not equal to democratic expertise • Paternalism: To do something good to a person agianst that persons will • Whistle-blowing: Ringing the alarm • Cooperation: Towards a common goal

WHISTLE-BLOWING • Conditions for justified whistle-blowing (De. George) • • • Considerable Harm Closest Supervisor Internal W-B Evidence that would convince an impartial observer Good reason to believe harm can be avoided • However, demands large sacrifices • Not always effective

THE RESPONSIBILITES OF COMPUTER SCIENTISTS: AN EXAMPLE From the Guardian 2017: How white engineers built racist code – and why it's dangerous for black people As facial recognition tools play a bigger role in fighting crime, inbuilt racial biases raise troubling questions about the systems that create them From the Independent 2018: • BITCOIN WILL USE 0. 5% OF WORLD’S ELECTRICITY BY END OF 2018, FINDS STUDY • A single bitcoin transaction requires as much electricity as the average Dutch household uses in a month